Multi-sensor distributed fusion estimation with applications in networked systems

Multimodal sentiment analysis is drawing an increasing amount of attention these days. It enables mining of opinions in video reviews which are now available aplenty on online platforms. However, multimodal sentiment analysis has only a few high-quality data sets annotated for training machine learning algorithms. These limited resources restrict the generalizability of models, where, for example, the unique characteristics of a few speakers (e.g., wearing glasses) may become a confounding factor for the sentiment classification task. In this paper, we propose a Select-Additive Learning (SAL) procedure that improves the generalizability of trained neural networks for multimodal sentiment analysis. In our experiments, we show that our SAL approach improves prediction accuracy significantly in all three modalities (verbal, acoustic, visual), as well as in their fusion. Our results show that SAL, even when trained on one dataset, achieves good generalization across two new test datasets.

Select-additive learning: Improving generalization in multimodal sentiment analysis

Select-Additive Learning: Improving Generalization in Multimodal Sentiment Analysis

Automatic Face Recognition (FR) presents a challenging task in the field of pattern recognition and despite the huge research in the past several decades; it still remains an open research problem. This is primarily due to the variability in the facial images, such as non-uniform illuminations, low resolution, occlusion, and/or variation in poses. Due to its non-intrusive nature, the FR is an attractive biometric modality and has gained a lot of attention in the biometric research community. Driven by the enormous number of potential application domains, many algorithms have been proposed for the FR. This paper presents an overview of the state-of-the-art FR algorithms, focusing their performances on publicly available databases. We highlight the conditions of the image databases with regard to the recognition rate of each approach. This is useful as a quick research overview and for practitioners as well to choose an algorithm for their specified FR application. To provide a comprehensive survey, the paper divides the FR algorithms into three categories: (1) intensity-based, (2) video-based, and (3) 3D based FR algorithms. In each category, the most commonly used algorithms and their performance is reported on standard face databases and a brief critical discussion is carried out.

A review on state-of-the-art face recognition approaches

Spatial-temporal pyramid based Convolutional Neural Network for action recognition

Region-based Mixture Models for human action recognition in low-resolution videos

Abrupt motion tracking via nearest neighbor field driven stochastic sampling

Video pose estimation with global motion cues

Recognizing human action from low-resolution (LR) videos is essential for many applications including large-scale video surveillance, sports video analysis and intelligent aerial vehicles. Currently, state-of-the-art performance in action recognition is achieved by the use of dense trajectories which are extracted by optical flow algorithms. However, the optical flow algorithms are far from perfect in LR videos. In addition, the spatial and temporal layout of features is a powerful cue for action discrimination. While, most existing methods encode the layout by previously segmenting body parts which is not feasible in LR videos. Addressing the problems, we adopt the Layered Elastic Motion Tracking (LEMT) method to extract a set of long-term motion trajectories and a long-term common shape from each video sequence, where the extracted trajectories are much denser than those of sparse interest points(SIPs); then we present a hybrid feature representation to integrate both of the shape and motion features; and finally we propose a Region-based Mixture Model (RMM) to be utilized for action classification. The RMM models the spatial layout of features without any needs of body parts segmentation. Experiments are conducted on two publicly available LR human action datasets. Among which, the UT-Tower dataset is very challenging because the average height of human figures is only about 20 pixels. The proposed approach attains near-perfect accuracy on both of the datasets.

/pdf/recognizing-human-actions-from-low-resolution-videos-by-31vm1t45tg.pdf

Recognizing human actions from low-resolution videos by region-based mixture models

This paper presents a real time multirate multisensor state fusion estimation algorithm, where data from multisensors are fused without interpolation or augmentation of state or measurement dimensions, in which different sensors may have different sampling rates and may encounter missing measurements. The state model is formulated at the highest sampling rate, and the measurement equations for each sensor are known. The generated state estimation algorithm is optimal in the sense of linear minimum variance, and the simulation on the estimation of the current of a simple two-coil transformer illustrate its' feasibility and effectiveness.

http://ieeexplore.ieee.org/iel5/5980617/6000362/06000856.pdf

Feng Lv

Papers

Region-based Mixture Models for human action recognition in low-resolution videos

Abrupt motion tracking via nearest neighbor field driven stochastic sampling

Video pose estimation with global motion cues

Recognizing human actions from low-resolution videos by region-based mixture models

Optimal multirate filtering with its application in estimation of the current of a transformer